Temporal difference approach to coordinated motion control of cooperating two link robots
To perform complex manufacturing operations, two or more manipulators are made to work in concert. When robots work independently of other robots, small errors made (e.g., due to inaccuracies in modeling of the manipulator) by individual robots may be acceptable. However, when robots work together, then high precision is required. This calls for the use of adaptive controllers in order to minimize errors. This paper discusses the application of Temporal Difference Learning (TDL) method, wherein stiffness of manipulator is adapted based on the feedback obtained from force/torque sensors. In order to accomplish this, simulation was carried out by adding feedback (force and change of force) to controllers, so that required trajectory could be adhered. Normally this error (deviation from required trajectory) occurs due to non-availability of the correct values of stiffness of the system. Stiffness of system is difficult to calculate due to inherent complexities in formulating an accurate dynamic model of system. Variation in parameters, for example change of friction due to aging, change of moment of inertia due to changes in payload position and orientation, significantly affect the dynamic model of manipulator. One of the ways to achieve compliance is by updating the dynamic model of the system. The other way is to use the external control loop which provide manipulator with set-points such that the desired compliance can be achieved. This paper demonstrates the appropriateness of TDL method in updating the dynamic model of the system. This updated model is then used to calculate the torques of the joints. As the process of learning converges, the function learned represents a nearly perfect model of the stiffness of the system.
American Society of Mechanical Engineers, Dynamic Systems and Control Division (Publication) DSC
Volume / Issue
Start / End Page